Abstract: Based on observations of daily temperature at 2052 ground stations and outputs of General Circulation Models (GCMs) from the Coupled Model Intercomparison Project Phase 5 (CMIP5), we study the changes in seasonal temperature extremes and their spatio-temporal patterns across China during the period from 2071 to 2100 relative to the period from 1971 to 2000. The grid-scale extremes estimated from GCM outputs are statistically downscaled to the site scale by transfer functions developed based on the quantiles of grid-scale simulations and on-site observations of the extremes. The assessment of the downscaled seasonal temperature extremes based on Taylor diagram and two-sample Kolmogorv-Smirnov test show that the seasonal maximum/minimum of daily maximum and minimum temperature can be well represented by the downscaled GCM outputs. The seasonal extremes of daily maximum and minimum temperature increase faster in autumn and winter compared to those in spring and summer. The warm nights defined as seasonal fraction of time with daily minimum temperature higher than its 90th percentile centered on a 5-day window are projected to increase substantially in summer. In general, temperature extremes in northern China increase faster than southern China, especially in autumn and winter. The largest increases in seasonal minimum are identified in the most northeastern China in autumn and winter, with magnitudes of >8 °C. In southern China, the increases of seasonal extremes of daily maximum and minimum temperature are generally < 2 °C under RCP2.6, and 4 °C under RCP8.5.